Function task

qa/tasks/lost_unfound.py:13–180

Test handling of lost objects. A pretty rigid cluster is brought up and tested by this task

(ctx, config)

log = logging.getLogger(__name__)

def task(ctx, config):
    """
    Test handling of lost objects.

    A pretty rigid cluster is brought up and tested by this task
    """
    POOL = 'unfound_pool'
    if config is None:
        config = {}
    assert isinstance(config, dict), \
        'lost_unfound task only accepts a dict for configuration'
    first_mon = teuthology.get_first_mon(ctx, config)
    (mon,) = ctx.cluster.only(first_mon).remotes.keys()

    manager = ceph_manager.CephManager(
        mon,
        ctx=ctx,
        logger=log.getChild('ceph_manager'),
        )

    while len(manager.get_osd_status()['up']) < 3:
        time.sleep(10)

    manager.wait_for_clean()

    manager.create_pool(POOL)

    # something that is always there
    dummyfile = '/etc/fstab'

    # take an osd out until the very end
    manager.kill_osd(2)
    manager.mark_down_osd(2)
    manager.mark_out_osd(2)

    # kludge to make sure they get a map
    rados(ctx, mon, ['-p', POOL, 'put', 'dummy', dummyfile])

    manager.flush_pg_stats([0, 1])
    manager.wait_for_recovery()

    # create old objects
    for f in range(1, 10):
        rados(ctx, mon, ['-p', POOL, 'put', 'existing_%d' % f, dummyfile])
        rados(ctx, mon, ['-p', POOL, 'put', 'existed_%d' % f, dummyfile])
        rados(ctx, mon, ['-p', POOL, 'rm', 'existed_%d' % f])

    # delay recovery, and make the pg log very long (to prevent backfill)
    manager.raw_cluster_cmd(
            'tell', 'osd.1',
            'injectargs',
            '--osd-recovery-delay-start 1000 --osd-min-pg-log-entries 100000000'
            )

    manager.kill_osd(0)
    manager.mark_down_osd(0)
    
    for f in range(1, 10):